Taxonomical modeling and classification in space hardware failure reporting.

Area of Science:

• Data Science
• Natural Language Processing
• Machine Learning

Background:

• Manual analysis of over 54,000 NASA space hardware failure reports is impractical for identifying trends.
• Expert knowledge is often unavailable or unstructured for large datasets.

Purpose of the Study:

• To develop an automated taxonomical model for text mining and knowledge extraction from engineering failure reports.
• To structure unstructured text data for trend identification and root cause analysis.

Main Methods:

• Utilized Latent Dirichlet Allocation (LDA) algorithms combined with Bidirectional Encoder Representations from Transformers (BERT).
• Developed an automated taxonomical model to structure text content.

Main Results:

• The combined LDA and BERT model provides structure to text content for analysis.
• Enables knowledge extraction and trend identification from large datasets.

Conclusions:

• Automated taxonomical models using LDA and BERT can overcome limitations of manual analysis and expert dependency.
• Facilitates efficient analysis of extensive engineering failure report datasets.